The Frankenbear Experiment: Looking for part-based similarity effects on search guidance with complex objects

Similarity is a key concept in many theories of visual search, but the effects of object similarity on search have only been explored using simple stimuli. Given that real-world objects can be similar and different in many respects, it is unclear whether relationships obtained for simple objects will generalize to more complex search stimuli. In this experiment, the heads, arms, legs or torsos of distractor teddy bear objects (all photorealistic images) were replaced with the equivalent parts from a target teddy bear. This similarity manipulation transplanted target features to the distractors on a part-by-part and pixel-by-pixel basis. Subjects were shown a target preview followed by a 4 or 8-object present/absent search display consisting of random, unaltered bears or bears that were manipulated to have one, two or three parts matched to the target. Trials were also either distractor heterogenous, where each distractor was a different bear with different parts matched to the target, or distractor homogenous, where distractors were the same bear with the same matched parts. Consistent with previous work (e.g., Duncan & Humphreys, 1989), we found that errors and RT slopes increased with target-distractor similarity and decreased with distractor-distractor similarity. Increasing target-distractor similarity also resulted in fewer initial saccades directed to the target and more distractors fixated before the target. However, similarity did not interact with distractor homogeneity/heterogeneity in our task (contrary to previous work), and oculomotor analyses revealed minimal effects of set size or distractor heterogeneity on overt search guidance. We interpret these patterns as evidence for a dissociation between distractor-distractor similarity and search guidance; homogenous distractors may speed manual search decisions and reduce errors, but we found little evidence suggesting that distractor-distractor similarity improves search guidance. We speculate that this may be due to a reduced ability to group visually complex real-world objects.